Association In Repetitive And Low-Mappability Regions

Why Region Properties Matter

A structural-variant association test can be calculated correctly and still be misleading if the calls entering it are unreliable. This risk is especially important in segmental duplications and other repetitive sequence, where reads from one genomic copy can align to another. A caller may then interpret misplaced reads, apparent read-depth changes, or misleading breakpoint evidence as a deletion, duplication, or other structural variant.

The key question for this tutorial is not simply whether a variant passes a general quality filter. It is whether the particular genomic region can support reliable, comparable calls in the samples being tested. Region properties affect read placement and therefore can affect call sensitivity, false-positive rates, and the fraction of samples that can be called at all.

How Repetitive Sequence Can Distort Calls

A segmental duplication is a genomic region that has a substantially similar copy elsewhere in the genome. Repetitive sequence can also occur in many copies, sometimes over shorter stretches. If reads from similar copies are difficult to distinguish, an aligner may place a read at the wrong copy, assign it low mapping quality, or leave it unmapped. Structural-variant callers use patterns such as read depth, discordant read pairs, and split reads; ambiguous placement can distort each of these signals.

• Read depth: reads from a similar copy may accumulate at the wrong locus, making apparent depth too high. Reads that fail to map may instead make depth appear too low. Either pattern can resemble a copy-number change.
• Breakpoint evidence: a read spanning a repeated sequence can be split or aligned across similar copies in a misleading way. The resulting apparent breakpoint may not correspond to a true rearrangement at that location.
• Discordant pairs: reads from different copies can be assigned to positions that imply an unexpected distance or orientation, resembling evidence for a structural variant.
• Reference limitations: missing, collapsed, or imperfectly represented sequence can make the alignment problem worse. A call can appear precise in genomic coordinates even when the underlying reads do not distinguish the locus reliably.

These problems do not mean that genuine structural variants cannot occur in repetitive regions. They mean that the evidence must be assessed in light of the region’s sequence properties. A high-quality-looking call from a difficult region is not automatically reliable, and a low call rate may reflect lack of information rather than absence of the variant.

Region-Specific Assessment Before Testing

Before association testing, assess the region that contains the candidate event, rather than assuming that a general variant-level filter resolves mapping ambiguity. A practical assessment brings together reference annotation, read-level evidence, and sample-level callability. No single statistic can replace these checks.

The criteria used in this process should be appropriate to the assay and specified without choosing them to obtain a desired association result. A fixed mapping-quality or unique-read threshold cannot be recommended for every platform and locus. If a region cannot distinguish copies well enough to support reliable calls, applying a more stringent version of an ordinary variant filter may not solve the underlying problem.

Why Differential Error Can Create an Association

A central concern is that call errors may differ between the groups being compared. For example, cases and controls may have been sequenced in different batches, or may differ in read quality. If a difficult region has more false-positive calls in one group, the observed carrier proportions can differ even when the true carrier proportions are equal.

Let \(p_g\) be the true carrier proportion in group \(g\), \(Se_g\) the probability that a true carrier is called positive, and \(f_g\) the probability that a true noncarrier is called positive. The observed positive-call proportion \(q_g\) follows by conditioning on true carrier status:

The first term is the proportion of the group that consists of true carriers who are detected; the second is the proportion of true noncarriers who receive false-positive calls. If the true carrier proportions are equal but the false-positive rates differ, the observed proportions can still differ. This is why comparable region-specific calling matters for association testing.

Callability Is Not the Same as Genotype

A genotype or dosage summarizes the inferred structural-variant state. Callability concerns whether the available data support assigning that state for a particular sample. In a repetitive region, some samples may have too little uniquely placed evidence for a confident call. Treating those samples as reference or noncarrier samples changes the analysis rather than merely handling missing information.

Callability should be examined separately by case-control group and, where relevant, by sequencing batch or other technical group. A similar overall call rate can conceal an imbalance within a subgroup, while a difference in call rate can signal that the groups are not being compared on equally informative data. If callability depends on phenotype group, a simple analysis restricted to called genotypes may also be biased; the reason for missing calls matters.

Safeguards and Their Limits

Safeguards should address the specific failure mode, not simply add more filtering. Useful checks include verifying that supporting reads are locus-specific, inspecting coverage and breakpoint evidence, and comparing these features across groups. When a region is important to the scientific question, independent evidence from an assay with different measurement properties can help establish whether the event is real. Any validation should target the same event and genomic locus, rather than merely confirming that a related copy exists.

For the association analysis, document how ambiguous or uncallable samples are handled and check that the rule is applied consistently. If cases and controls were processed in separate batches, region-level callability and evidence should be examined within those batches as well as overall. Statistical adjustment for a batch label is not a substitute for reliable measurement when batch and phenotype group are inseparable.

Possible decisions include retaining a well-supported event with adequate, comparable calls; obtaining additional locus-specific validation; or excluding the region from the association analysis. Exclusion can be scientifically preferable to reporting a precise estimate from calls that do not reliably identify the locus. The decision should reflect the evidence and analysis plan, not the direction or significance of the observed association.

Common Mistake

• Wrong reasoning: “The variant passed the caller’s quality filter, so its genotypes are reliable enough to test.”
• Correction: A caller’s filter does not necessarily resolve ambiguity caused by similar genomic copies. Assess whether the reads identify the locus, whether samples are comparably callable, and whether the evidence is consistent across the groups being compared.
• Wrong reasoning: “A sample without a call is a noncarrier.”
• Correction: A missing or uncallable genotype is not evidence of the reference state. Treating it as a noncarrier can distort carrier proportions, particularly when callability differs by group.